36 views

Uploaded by Mohammad Saif Uddin

Cash flows

- 1. Forecasting Introduction
- WINSEM2018-19_MEE1014_TH_MB102_VL2018195002165_Reference Material I_Unit 1.2_IEM (1).pdf
- bb sbi FAN-MANAGING EARNING USING CLASIFICATION SHIFTING.pdf
- 17833356
- Sales Forecasting
- Evaluation of the Quality of a Prognosis for an Industrial Product using the Regression Analysis
- SupplyChainPlanningAndControl BA En
- chap_15.ppt
- Gat Subject Management Sciences Management Mcqs1 50
- Forecasting
- Operations Management chapter 3 - Forecasting
- Demand Forecasting
- 4 Cast Demand
- Forecasting
- Solanib_Quanti.pdf
- Swanson & Tedrow Paa 2016 v1
- Demand Forecasting Techniques
- 360 case 2 full
- Demand and Supply Network Planning in SCM.pdf
- 170

You are on page 1of 49

Dan Givoly Smeal College of Business The Pennsylvania State University e-mail: dgivoly@psu.edu Carla Hayn Anderson Graduate School of Management University of California at Los Angeles e-mail: carla.hayn@anderson.ucla.edu and Reuven Lehavy Ross School of Business University of Michigan e-mail: rlehavy@umich.edu

May 2008

We gratefully acknowledge helpful comments by Guojin Gong, Bin Ke, Karl Muller, Charles Wasley, Joanna Wu, Jerry Zimmerman and workshop participants at Penn State University and the University of Rochester.

This paper examines properties of analysts cash flow forecasts and compares these properties with those exhibited by analysts earnings forecasts. Our results indicate that analysts cash flow forecasts are of a considerable lower quality than their earnings forecasts. They are less accurate and improve at a slower rate during the forecast period. Further, analysts cash flow forecasts appear to be, in essence, a nave extension of their earnings forecasts and provide no incremental information on expected changes in firms working capital. Consistent with their low quality and in contrast to their earnings forecasts, analysts forecasts of cash flows are of limited information content and are only weakly associated with stock price movements. Finally, a measure of expected accruals based on the difference between analysts earnings and cash flow forecasts has a very low power in detecting earnings management.

1. Introduction Financial analysts generate a number of important products, among them earnings forecasts, stock recommendations and target stock prices. In recent years, analysts have gradually introduced yet another product -- forecasts of firms operating cash flow. The relative frequency of firms for which cash flow forecasts are provided in addition to earnings forecasts has increased from 2.5% in 1993 to over 55% in 2005. This trend along with the greater availability of analysts cash flow forecasts through commercial databases have led to an increase in the number of studies that employ these forecasts in a variety of research settings. While the accounting and finance literature has extensively examined various properties of analysts earnings forecasts, very little is known about the properties of analysts cash flow forecasts, possibly due to their short history and because they are still not as widely available as earnings forecasts. In this paper, we attempt to close this gap in knowledge by exploring basic forecasting properties such as accuracy, bias and efficiency, and by benchmarking the performance of analysts cash flow forecasts against that of the well-studied earnings forecasts. In addition to examining basic forecasting properties, we investigate the extent of sophistication reflected in analysts cash flow forecasts. Specifically, we examine whether these forecasts incorporate projections of working capital accruals or merely represent the addition of some estimate of depreciation and amortization to the already-produced earnings forecasts. We also evaluate two potential uses of analysts cash flow forecasts in research settings. The first is as a proxy for the unobservable market expectation of cash flows. In line with past research on analysts earnings forecasts, we gauge the extent to which analysts cash flow forecasts represent the market expectation by the association between their forecast errors and stock returns. The second potential use of cash flow forecasts in a research context that we consider is how well they serve as a basis for estimating expected accruals and extracting unexpected accruals. Expected accruals can be inferred by subtracting analysts cash flow

forecasts from their contemporaneous earnings forecasts. We test the power and efficiency of this accrual model in detecting earnings management and compare the results to those produced by the modified Jones Model (Dechow, Sloan and Sweeney, 1995) and Dechow and Dichev (2002) models commonly used in the literature. This examination is related to, and helps shed light on, another question: Do analysts anticipate earnings management? If analysts anticipate earnings management, their earnings forecasts will reflect the accruals used to manage earnings. The difference between their earnings forecasts and cash flow forecasts would be correlated with the presence of earnings management. Note that testing whether an accruals model based on analysts forecasts of both cash flows and earnings is effective in detecting earnings management constitutes a joint test of the quality of analysts cash flow forecasts and analysts inability or unwillingness to exclude the effect of anticipated earnings management from their earnings forecasts. Our results indicate that analysts cash flow forecasts are of a considerable lower quality than their earnings forecasts. Cash flow forecasts are not only less accurate but further, the rate of their improvement over the forecast period is much lower than that of earnings forecasts. The lower accuracy of cash flow forecasts relative to earnings forecasts is only partially attributable to data quality issues or the inherent difficulty in forecasting cash flows stemming from the higher variability of firms cash flow series relative to their earnings series. Our results further indicate that analysts cash flow forecasts are in essence a nave extension of their earnings forecasts. As a result, they provide no incremental information on expected changes in firms working capital. Possibly as a result of their low quality and data measurement errors, analysts cash flow forecasts, in contrast to their earnings forecasts, do not appear to be a better proxy for the unobservable market expectation of future cash flows than do mechanical time series models. Two additional findings of our paper are relevant for future research that uses analysts cash flow forecasts. First, these forecasts, in contrast to analysts earnings forecasts, are only weakly associated with stock price movements. Thus, in a research setting these forecasts are not

a good surrogate for the unobservable market expectations of cash flows. The second finding, that the estimation of expected accruals based on the difference between analysts earnings and cash flow forecasts is not effective in detecting earnings management, is relevant for future research on earnings management. The paper contributes to the accounting literature in several ways. It is the first to comprehensively document important properties of cash flow forecasts and to assess their quality. The finding of a low accuracy of these forecasts, the fact that they represent in essence a trivial extension of analysts earnings forecasts and the evidence of serious data quality issues are relevant for investors who might consider these forecasts in valuation and investment decisions and researchers using cash flow forecasts in different research contexts. The paper also provides two results relevant for future research. One is that errors in analysts cash flow are only marginally associated with stock price movements after controlling for analysts earnings forecasts. The other is that analysts earnings and cash flow forecasts can serve to generate an accrual expectation model superior to those commonly used in the literature to detect earnings management. The remainder of the paper is organized as follows. In the next section we review the related research on cash flow forecasts. The empirical design and tests used to explore the properties and attributes of cash flow forecasts are contained in the third section. Data used in the tests are described in section 4. The results are presented and discussed in section 5. The last section of the paper contains summary and concluding remarks. 2. Related Research The first paper that documents the increased propensity of analysts to issue cash flow forecasts and analyzes the explanation for this trend is DeFond and Hung (2003). They hypothesize and provide empirical support for the notion that the increased frequency of cash flow forecasts is related to demand by investors who are increasingly concerned about the inherent shortcomings of accrual accounting, such as its subjective nature and its susceptibility to

earnings management. Since cash flow from operations is perceived to be more objective and less vulnerable to management manipulation, it is commonly viewed as a valuable supplement to earnings information. DeFond and Hung (2003) show that, consistent with cash flow forecasts being driven by investors demand, analysts propensity to produce these forecasts increases with the magnitude of accruals, managerial latitude in choosing accounting methods, earnings volatility, capital intensity and financial distress. In a complementary study, DeFond and Hung (2007) examine analysts propensity to issue cash flows across countries with different reporting regimes. They hypothesize and find that analysts are more likely to supplement their earnings forecasts with cash flow forecasts in countries where investor protection is poor and earnings are of a lower quality.1 Ertimur and Stubben (2005) show that the supply of cash flow forecasts, in addition to being affected by demand factors, is also influenced by analysts and brokerage characteristics.2 In line with the notion that cash flow forecasts are driven by investor demand arising from earnings quality concerns, a number of studies examine how the presence of analysts forecasts of cash flows affects earnings quality, predictability and valuation. McInnis and Collins (2006) find that the presence of a cash flow forecast elevates earnings quality. They attribute this result to the fact that cash flow forecasts implicitly provide a forecast of the accrual portion of the earnings forecast, thereby increasing the transparency of any accrual manipulations that might be undertaken to meet earnings thresholds. While cash flow forecasts might reduce earnings management, Call (2007) finds that their presence increases the propensity of firms to manipulate their cash flow from operations. Consistent with the signaling value of the presence of cash flow forecasts, Call (2007) finds that investors assign more weight to the cash flow component of earnings in stock

Hail (2007) concludes that the DeFond and Hung (2007) findings which are based on cross-sectional analyses do not necessarily hold in a time-series setting. 2 Investor demand is among the factors found also to prompt management cash flow forecasts (see Wasley and Wu, 2005).

valuation when cash flow forecasts are provided. Pae, Wang and Yoo (2007) find that analysts earnings forecasts are more accurate if they also produce a cash flow forecast. Other studies use analysts cash flow forecasts to proxy for investors expectation of cash flows. DeFond and Hung (2003) use these forecasts to estimate the unexpected cash flows in their analysis of earnings announcement period returns. Similarly, McInnis and Collins (2006) use cash flow forecasts to derive unexpected cash flows in their analysis of the differential response of investors to cash flow and accrual information provided in the earnings announcement. Melendrez, Schwartz and Trombley (2005) compare the market valuation of the unexpected accrual component to that of the unexpected cash flow component in reported earnings. To make this comparison they construct a measure of unexpected accruals based on analysts cash flow forecasts. Finally, the errors in analysts cash flow forecasts are used by Zhang (2008) to examine the market reward and analysts response to meeting or beating these forecasts, and by Brown and Pinnelo (2008) to analyze the circumstances present when firms meet one of the thresholds earnings or cash flow forecasts but not the other. As this literature review indicates, some studies have used the values of the cash flow forecasts in their design while other studies used merely the presence of cash flow forecasts as an indicator variable. The results of our study regarding the quality of cash flow forecasts have implications for both types of studies. 3. Empirical Design and Tests 3.1. Forecasting performance We examine four properties of the performance of analysts forecasts of cash flows accuracy, bias, efficiency and intra-year improvement. We then compare these properties of analysts cash flow forecasts to those of analysts earnings forecasts. We also compare analysts ability to forecast cash flows to that of a mechanical, time-series model. We separately examine cash flow forecasts for year t made at two points in time at the beginning of the year soon after

the announcement of the prior years (year t-1) earnings and at the end of the year immediately before the announcement of the current years (year t) earnings. Accuracy is assessed using two forecast error measures, the relative absolute forecast error, AbError, and the relative squared forecast error, SqError, as follows (the subscripts i and t are used to indicate the firm and year, respectively, throughout the paper): Relative Absolute Error = AbErrorit = Ait Fit / Ait Relative Squared Error = SqErrorit = (Ait Fit)2 / Ait (1) (2)

where A is the actual value (of earnings per share or cash flows per share) and F is the forecasted value. 3 Bias is measured as the signed forecast error, namely: BIASit = (Ait Fit) / Ait (3)

Forecast efficiency is gauged by the serial correlation in the prediction error. Serial correlation is measured by the slope coefficient, , of the following time-series regression which relates the cash flow forecast error in the current year to that in the prior year: FEit = + FEit-1 + it (4)

The forecast error, FE, is measured as the actual value minus the forecasted value, alternately standardized by the absolute value of the actual value or the beginning-of-the-period stock price. The presence of a significant correlation would indicate that information contained in the prediction error is not used efficiently in generating future forecasts. Intra-year improvement in the forecasts is measured as the rate of decline in the analysts forecast error (as defined in the preceding paragraph) between their beginning-of-year and endof-year forecasts. Specifically: IMPROVEMENTit = 1 - (FEi,END(t) / FEi,BEG(t) ) (5)

These four properties of analysts forecasts of cash flows are compared to those of analysts earnings forecasts and, with respect to the first three properties discussed above, a time-

We replicated all of the analyses of the error measures using price, rather than the absolute value of the actual numbers (cash flow or earnings), as a deflator. There was very little difference in the results and the inferences.

series model proposed in the literature. This model relies on past cash flows and the reversal of past accruals. As shown by Dechow, Kothari and Watts (1998), past earnings are a better predictor of future cash flows than are past cash flows. Extending this work, Barth, Cram and Nelson (2001) find that disaggregating past cash flows and accruals into components improves cash flow predictability. We follow their approach which predicts cash flow from operations (CF) in year t as a function of the previous years cash flow from operations, changes in working capital accounts (accounts receivable (AR), inventory (INV), and accounts payable (AP)), the level of depreciation (DEP) and other accruals (Other) as follows:4 CFit= 0 + 1CFit-1+2AR it-1 + 3INV it-1 + 4AP it-1+5DEP it-1+6Other it-1 + it 3.2. Sophistication of analysts forecasts of cash flow A simple relation exists between earnings and cash flow from operations. Specifically, cash flows from operations is equal to earnings plus depreciation and amortization (and other non-cash charges) minus (plus) non-cash gains (losses) and the net increase (decrease) in working capital accruals. Since analysts earnings forecasts routinely accompany their cash flow forecasts, a question arises regarding the incremental contribution of the latter. In particular, do analysts cash flow forecasts merely represent the analysts earnings forecasts adjusted for projected depreciation and amortization or are they more sophisticated as a result of incorporating the more challenging prediction of working capital accruals? A review of scores of analysts research reports indicates that although some appear to have incorporated working capital accruals in their cash flow forecasts, most do not spell out how they define cash flows nor do they detail the derivation of their cash flow forecasts. The underlying assumption of most studies that use analysts cash flow forecasts either as proxy for market expectations or as a signal for earnings quality is that these forecasts are more sophisticated than a simple derivative of analysts own earnings forecasts. We address whether cash flow forecasts are indeed (6)

Other is computed as income from continuing operations (INCCO) minus the sum of cash flow from operations, the change in the three working capital accounts and depreciation (Otherit-1= INCCOit-1 (CFit-1 + ARit-1 + INVit-1 + APit-1 + DEPit-1)).

sophisticated or take a more nave form such as a straightforward extension of analysts earnings forecasts is an empirical issue.5 We assess the extent of sophistication inherent in analysts cash flow forecasts through a number of related tests. First, we examine the incremental accuracy of analysts forecasts of cash flows over a cash flow prediction model that represents a nave extrapolation of analysts own earnings forecasts. Specifically, we model analysts forecasts of next years cash flows , F(CF), as a simple addition of their contemporaneous forecast of earnings for the next year, F(Earnings), plus the current years realized depreciation and amortization expense, DEP, as follows (the subscripts i and t denote the firm and year, respectively): F(CFit) = 0 + 1F(Earningsit) + 2DEP it + it (7)

Standardizing the error terms by, alternatively, the actual cash flow or the firms stock price, we compare the magnitude of the relative absolute error and the relative squared error with the analysts cash flow error measures in (1) and (2), respectively. The second test of the extent of sophistication in analysts cash flow forecasts is based on the correlation between the error in analysts earnings forecasts and the error in their cash flow forecasts. If analysts cash flow forecasts are merely a nave extrapolation of their earnings forecasts, we would observe a high correlation between the respective analysts forecast errors of cash flows and earnings. The third test of the extent of sophistication in analysts cash flow forecasts is based on the following regression: F(CFit) = 0 + 1 F(Earningsit) + 2DEPit + 3WCit + 4Otherit + it (8)

where F(CF) and F(Earnings) are the analysts forecasts for, respectively, cash flows and earnings for that year, DEPt , WCt and Othert are, respectively, the forecasted values for year t of the depreciation and amortization expense, change in the working capital accounts (Accounts

DeFond and Hung (2003, p. 84) conclude from several full-text reports by I/B/E/S-contributing analysts that it appears that the cash flow forecasts are not a trivial translation of predicted earnings, but rather the result of difficult and costly information processing that involves the prediction of working capital and deferred taxes. Both McInnis and Collins (2006, p.5) and Pae, Wang and Yoo (2007, p. 7) appear to rely on Defond and Hungs (2003) conclusion to motivate their use of analysts cash flow forecasts.

5

Receivable, Inventory and Accounts Payable), and all other adjustments needed to reconcile income from continuing operations in the forecast year with cash flow from operations. Since the forecasted values of DEPt , WCt and Othert are not observable, we assume perfect foresight by analysts and use the actual values of these variables for year t.6 If analysts form their cash flow forecasts by adding depreciation to their earnings forecasts and then subtracting (or adding) the net increase (decrease) in working capital, the slope coefficients i should not be significantly different from one. If, however, analysts fail to add back depreciation, adjust for the change in working capital accounts or make the other adjustments, the slope coefficients 2, 3 and 4 would not be significantly different from zero. 3.3. Analysts forecasts of cash flows as a surrogate for market expectations Analysts forecasts of earnings have been found to be of information content (Givoly and Lakonishok, 1984) and to be a better surrogate for the unobservable market expectations of earnings than forecasts produced by mechanical earnings prediction models (Fried and Givoly, 1982; Brown, Hagerman and Zmijewski, 1987). To gauge the extent to which analysts cash flow forecasts provide a reasonable surrogate for investors expectations, we estimate the association between abnormal returns and unexpected cash flows as proxied by the prediction error of analysts cash flow forecasts. Because the prediction error from analysts cash flow forecasts may be correlated with the error from analysts earnings forecasts, the positive correlation between the errors of the contemporaneous earnings and cash flow forecasts may create a spurious correlation between the abnormal returns and cash flow forecast errors even though analysts cash flow forecasts may not have incremental information content beyond that provided by earnings forecasts. To control for the correlation between the errors of the earnings and cash flow forecasts, we estimate the incremental association between abnormal returns over the forecasted year t and unexpected cash flows for this year, controlling for unexpected earnings, using the following regression: CARjt = +1FE(Earningsit) + 2FE(CFit) + it

6

(9)

Using the actual values of these three variables in year t-1 in the estimation of regression (8) yields similar results.

10

where CAR is the cumulative abnormal return over the 12-month period beginning four months after the end of fiscal year t-1and ending the third month following the end of fiscal year t.7 FE (Earnings) and FE(CF) are, respectively the beginning-of-the-year forecast errors for earnings and cash flows, respectively, deflated by price. We also examine the association between the earnings announcement period returns and the errors in the forecasts of cash flow and earnings outstanding at the time of the announcement by estimating regression (9). The earnings announcement period consists of the four-day window beginning one day prior to the earnings announcement. We further assess the incremental information content of cash flow forecasts with respect to the mechanical cash flow prediction model as specified in equation (6) above. This assessment is done by adding the prediction error from that mechanical model as an explanatory variable in regression (9). 3.4. Analysts forecasts of cash flows as an accruals expectation model Recent accounting research, primarily studies on earnings management, has been challenged by the daunting task of measuring what has been termed abnormal accruals. A number of expected accruals models have emerged as the gold standards of the literature, most notably the Jones model (1991), the modified Jones model (as proposed by Dechow, Sloan and Sweeney, 1995) and the Dechow and Dichev (2002) model. Even though widely used in the earnings management research, these accruals expectation models suffer from various limitations. For example, all of the models make certain assumptions about the functional relationship between accruals and activity measures such as the change in sales or the level of plant, property and equipment that, while plausible, may or may not strictly hold. The models further assume that the relationship between cash flows and accruals is linear, thus ignoring the

To calculate monthly abnormal returns over the forecast year (year t), the market model is estimated over the 60month period ending with the fiscal yearend of year t-1. The value-weighted index is used to proxy for the market return. Regression parameters from the market model are then used to calculate monthly abnormal returns. The cumulative abnormal return, CAR, for year t is the sum of the twelve monthly abnormal returns beginning with the fourth month of fiscal year t in order to exclude the effect of the announcement of earnings for year t-1. Very similar results are obtained when size-adjusted returns are used.

11

asymmetry in the gain and loss recognition of accruals (see Ball and Shivakumar (2005)). Indeed, these models are far from perfect in detecting earnings management (as noted, for example, by Dechow, Sloan and Sweeney (1995)). Analysts cash flow forecasts coupled with analysts earnings forecasts implicitly provide an accrual expectation model since the difference between these forecasts can be viewed as the forecasted (expected) accruals. The difference between actual accruals and this expectation is thus a measure of unexpected accruals. We test the power and efficiency of this analysts-based accrual expectation model in detecting earnings management by comparing it to the performance of the modified Jones and Dechow and Dichev (2002) models. Specifically, we estimate the modified Jones model from the following regression estimated cross-sectionally within each two-digit SIC code industry (the subscripts i and t denote the firm and time, respectively): TACCit/TAit-1 = a0 + a1(1/TAit-1) + a2[(REVit - TRit ) / TAit-1] + a3(PPEit / TAit-1) + it (10) where TACC is the firms total accruals, defined as the difference between income from continuing operations and cash flows from operating activities adjusted for extraordinary items and discontinued operations. REV is the change in the firms revenues, TR is the change in trade receivables, and PPE is the amount of gross property, plant and equipment. All variables are standardized by total assets at the beginning of the year, TAt-1. The Dechow and Dichev (2002) model of expected accruals is estimated from the following regression: TCACCit/AvTAit = 0 +1CFOit-1/AvTAit + 2CFOit/AvTAit + 3CFOii+1/AvTAit + it (11)

where TCACC is total current accruals, CFO is cash flows from operations, measured as income from continuing operations less total accruals, and AvTA is average total assets.8,9

Total current accruals is computed as operating assets (current assets excluding cash and short-term investments) minus operating liabilities (current liabilities excluding the current portion of long-term debt). 9 This model is not, strictly speaking, a prediction model since one of its predictors is the value of the next periods cash flows from operations. Nonetheless the model can still be used to detect earnings management on an ex post basis (e.g., by researchers or regulators).

12

To conduct these tests, we pre-identify two samples containing cases (firm-years) where earnings management is likely to have occurred. The first sample consists of firm-years for which earnings are eventually restated downward. For this restatement sample, the presumption is that the originally reported earnings were managed (hence the need for a restatement). When more than one year is restated for any given firm, we use only the first year in the sequence of restated periods both because the earnings management motivation is likely to be the strongest in this period and because subsequent periods accruals may reflect the reversal of any earnings management in the initial period. The second sample of cases where earnings management is likely to have occurred consists of years in which firms met or barely surpassed an earnings threshold. Two earnings thresholds are considered loss avoidance and avoidance of an earnings decline relative to the same quarter in the previous year. Earnings are identified as meeting or being just-above these thresholds when they exceed the thresholds by no more than k% of the end-of-quarter market values of equity where k is, alternately, equal to 0.25%, 0.50% and 1.0%. These cases are denoted as loss avoiders or earnings decline avoiders.10 To gauge the efficacy of the implied accrual expectation model derived from analysts cash flow forecasts, we compare the ability of this model to identify earnings management cases in the two samples described above with that of the modified Jones and Dechow and Dichev (2002) models. Specifically, for each of these models we derive unexpected accruals for every firm-year in the two samples. Abnormal accruals are identified as those whose standardized absolute value (the absolute value divided by the standard deviation of the unexpected accruals produced by the respective model across all firm-years) exceeds K (where K takes on the value 1.0, 1.5 or 2.0). Finally we tally the frequency of type I and type II errors of the three models in classifying firm-years as containing or not containing earnings management.11

10

The results tabulated in the paper are those obtained using a k of 1%. Using the lower values of k reduces considerably the number of earnings management cases but leaves the results intact. 11 Another procedure for testing the power of alternative accruals models to detect earnings management is a simulation analysis (e.g., Kothari, Leone and Wasley, 2005). This procedure is not applicable to our analysis since,

13

Note that the interpretation of the results concerning the power of the analysts-based accrual expectation model to detect earnings management depends on whether analysts anticipate earnings management and incorporate it in their earnings forecasts. If analysts do incorporate earnings management in their earnings forecasts, the measure of expected accruals derived from the difference between analysts cash flow and earnings forecasts would already reflect earnings management. Therefore, unexpected accruals derived from the model as described above would not necessarily indicate earnings management. Testing the power of the analysts-based accruals expectation model is thus a joint test of the power of the model and the assumption that analysts do not anticipate earnings management (or in some other manner incorporate earnings management in their earnings forecasts).12 4. Data and Sample One-year-ahead forecasts of cash flows and earnings as well as their respective actual values were obtained from the I/B/E/S detail file which contains the individual forecasts made by analysts (in contrast to the I/B/E/S Summary file which provides monthly average forecasts).13 For each fiscal year, we construct two measures of, separately, the consensus cash flow and the consensus earnings forecasts one based on the forecasts outstanding at the beginning of the year and the other based on the forecasts outstanding at the end of the year. To avoid stale forecasts, forecasts outstanding more than 90 days from the date of issuance are excluded. The beginning-of-year consensus forecasts are computed as the mean of all individual forecasts outstanding on the 30th day subsequent to the prior years earnings announcement. The end-ofunlike the predictions arising from mechanical models, analysts forecasts are not data-driven and thus, are unaffected by simulated alterations of the accrual data. 12 Past research provides conflicting evidence on whether analysts incorporate anticipated earnings management in their forecasts. Findings of Givoly, Hayn and Yoder (2007), Liu (2005) and Burgstahler and Eames (2003) suggest that analysts do incorporate the earnings management component in their earnings forecasts. In contrast, Abarbanell and Lehavy (2003a, 2003b) conclude that analysts either do not anticipate earnings management or they choose to exclude the managed earnings component from their forecasts. Ettredge, Shane and Smith (1995) find that analysts only partially discount overstated earnings in revising their earnings expectations. 13 Thomson Financial Glossary (2004), part of the Guide to Understanding Thomson Financial Terms and Conventions for the First Call and I/B/E/S Estimates Databases, defines cash flow per share forecasts as: cash flow from operations, before investing and financing activities, divided by the weighted average number of common shares outstanding for the year.

14

year consensus forecasts are computed as the mean of all individual forecasts outstanding when the current year earnings announcement is made. Actual values provided by I/B/E/S are used to calculate earnings and cash flow forecast errors to assure comparability with the forecasts.14 An issue identified by prior research (see, e.g., Melendrez, Schwartz and Tombley (2005)) is that, unlike its provision of an actual earnings number, I/B/E/S does not provide an actual value for cash flows for a large number of observations. For these observations (which represent more than half of the sample as shown later in table 4), we use the actual values reported by Compustat (annual data item #308 divided by the number of shares outstanding from I/B/E/S). Other financial statement information is retrieved from Compustat. Return data are obtained from the Center for Research in Security Prices (CRSP). 5. Results 5.1. Cash flow forecasts descriptive statistics Table 1 provides descriptive statistics on the availability of cash flow forecasts. Panel A of the table shows that the frequency of cash flow forecasts increased sharply over time, from 2.5% in 1993 to 57.2% in 2005. Mirroring this, the mean number of forecasts provided per firm has also increased over the period, from 3.2 to 14.3. Even accounting for this increase, for the most recent year presented, the average number of cash flow forecasts per firm is still about half of the average number of earnings forecasts per firm, 28.0. Panel B shows the frequency of cash flow and earnings forecasts across twelve industry groups. Considering the full sample period from 1993-2005, there appears to be great dispersion across industry groups in the frequency of cash flow forecasts. Cash flow forecasts are quite

14

In discussing the actual (reported) earnings data, Thomson Financial Glossary (2004), states that: Reported earnings are entered into the database on the same basis as analysts forecasts. By and large, this means operating earnings as opposed to net income...with very few exceptions analysts make their earnings forecasts on a continuing operations basis. This means that Thomson Financial receives an analysts forecast... after discontinued operations, extra-ordinary charges, and other non-operating items have been backed outThomson Financial adjusts reported earnings to match analysts forecasts on both an annual and quarterly basis. This is why Thomson Financial actuals may not agree with other published actuals; i.e. Compustat. While no explanation is provided about adjustments made to the reported cash flow amounts, this explanation likely applies also to the actual cash flow data.

15

prevalent in the Energy sector (issued for 78.7% of the firms) and much less common in other sectors such as Finance, Technology or Health Care. These results are consistent with those of DeFond and Hung (2003). However, by 2005, while almost all firms in the Energy sector received cash flow forecasts (96.8%), nine of the twelve industry groups had at least 50% coverage. The fairly monotonic increase in the frequency of analysts cash flow forecasts over time likely represents not only increased demand (particularly in recent years due to the wave of accounting shenanigans in the early 2000s) but also greater collection efforts of such forecasts by I/B/E/S. 5.2. Forecasting performance 5.2.1. Accuracy Table 2 presents the results on the accuracy of analysts cash flow forecasts as compared with the accuracy of their earnings forecasts, as well as with the accuracy of the Barth et al. model (see equation (6) above). As the table indicates, analysts cash flow forecasts are less accurate than their earnings forecasts. For example, the mean (median) relative squared error of analysts cash flow forecasts made at the beginning of the year is 0.8900 (0.1504) compared with only 0.3595 (0.0355) for analysts earnings forecasts. The same pattern is observed when relative absolute errors of analysts forecasts are considered. These differences are significant at the 1% significance level (with the exception of the mean relative absolute error which is significant at the 10% significance level). Panel B of table 2 shows the relative accuracy results for the end-of-year forecasts. Note that analysts forecasts of both cash flows and earnings are more accurate at year-end than at the beginning of the year, which is expected given the additional information available in the interim. However, cash flow forecasts made at year-end are still significantly less accurate than their contemporaneous earnings forecasts. The mean (median) relative squared error of analysts cash flow forecasts is 0.7604 (0.0797), considerably larger than the comparable statistics for analysts earnings forecasts, 0.0415 (0.0013). Comparable results are obtained for the relative absolute error. All differences are significant at the 1% significance level.

16

While analysts cash flow forecasts are less accurate than their earnings forecasts, they are generally more accurate than a mechanical time series model. Note that for both the beginning-of-year and end-of year forecasts shown in panels A and B, respectively, the results reported in this table show that analysts forecasts of cash flows are more accurate than the forecasts from the time-series model. This is particularly true for the end-of- year forecasts where all of the differences are statistically significant (see the last line of panel B). 5.2.1.1. Effect of learning and self-selection on accuracy One explanation for the average lower accuracy of analysts cash flow forecasts relative to their earnings forecasts is that cash flow forecasts are a relatively recent product that required a learning period by analysts. The average low accuracy of these forecasts may thus reflect the inaccuracy of the early periods forecasts and obscure the greater accuracy of cash flow forecasts produced in more recent years. Another explanation for the low accuracy of cash flow forecasts is that these forecasts are more likely demanded (and supplied) in cases where the forecasting of cash flow is more difficult. If this self selection explanation holds, we would expect this effect to be more pronounced in the early years when cash flow forecasts were available for relatively few firms (presumably those for which cash flow forecasting was most challenging) and less pronounced in recent years when these forecasts are much more widespread. To test the validity of both of these explanations, we analyze the accuracy results by years. The results, reported in panel C of table 2 do not support either of these explanations. As panel C shows, the accuracy of analysts cash flow forecasts has actually declined over time. The accuracy of analysts earnings forecasts has improved over the same period, making the superior accuracy of earnings even more pronounced in recent years. 5.2.1.2. Variability of the cash flow and earnings series One possible explanation for the higher error associated with cash flow predictions relative to that associated with earnings forecasts is the greater inherent variability of the cash flow series. We examine this explanation by comparing the variability of the two time-series in

17

the data. Our examination is based on a sample of firms that have at least 10 yeas of data in the 1993-2005 sample period. For each firm, we compute the variance of its cash flow and earnings distribution deflated, alternatively, by assets and price at yearend. As shown in table 3, the variance of the cash flow series is generally larger than that of the earnings series. For 56.8% (65.6%) of the firms, the variance of the cash flow series is higher than the variance of the earnings series when the series are scaled by assets (price).15 When scaled by assets, the median ratio between the cash flow and earnings variance, computed from the time-series of the 1,533 firms is 1.18 (with an inter-quartile range of 0.56 to 2.56). When price is used as the deflator, the median is 1.81 (with an inter-quartile range of 0.63 to 4.44). The mean values of the ratio, 5.69 and 8.62 when deflated by assets and price, respectively, are considerably higher than the median values reflecting the presence of extreme values in the cash flow series. The greater variability of the actual cash flow series, however, cannot fully explain the lower accuracy of the cash flow forecasts. Note that while the median ratio of the variance of the cash flow series to the variance of the earnings series is around 1.18, the squared forecast error in forecasting cash flows is larger than three times the earnings forecast error.16 Specifically, the ratio of the median square errors is close to 5 for the beginning-of-the-years forecasts (0.1504/0.0355, see panel A of table 2) and about 61 (0.0797/0.0013, see panel B of table 2) for the end-of-year forecasts. The above conclusion is reinforced when we estimate the regression where the dependent variable is the ratio of the cash flow squared forecast error to the earnings squared forecast error and the explanatory variable is the ratio of the variances of the cash flow series to the variance of the earnings series. The (untabulated) results based on over 3,000 firm observations show a

15

For the sample firms, the variance of the cash flow and earnings series are highly correlated. The Spearman correlation coefficients are 0.71 and 0.74, respectively, when scaled by assets and price; comparable Pearson correlation coefficients are 0.56 and 0.38. 16 The predictability of a variable and its variability are positively related. Consider for example an autoregressive behavior of earnings over time of the form: Et = + *Et-1 + et, where E is the earnings variable. Var(e) can be viewed as a predictability measure. Assuming further that the variance of earnings is stationary over time, earnings variability and predictability are related. Specifically, Var(e) = Var(E)* (1- 2). Dichev and Tang (2007) elaborate further on this relation.

18

significant slope coefficient suggesting that the higher cash flow forecast errors are explained in part by the higher variability of the cash flow series. Yet, the results also show a significant positive slope coefficient suggesting that the larger variability of the cash flow series falls short of fully explaining the lower accuracy of cash flow forecasts. We next turn to the relative measurement errors in the forecasted values and actual values of earnings and cash flows as a possible explanation for the differential accuracy of these series forecasts. 5.2.1.3. Effect of data quality on accuracy Defining the forecast error for earnings or cash flows requires that the forecast and the actual, or realized, values conform to the same definition. The realized reported values are the earnings or cash flows calculated in accordance with their Generally Accepted Accounting Principles (GAAP) definition. However, analysts forecasts do not always conform to their GAAP definition of the variable. For example, analysts often exclude from their earnings forecasts certain items that are considered transitory or otherwise unrelated to the core earnings of the company. Likewise, cash flow forecasts may exclude certain cash flow items that are included in cash flows from operating activities under SFAS 95. 17 In reporting the actual cash flows, I/B/E/S adheres to the same rule that it applies in reporting actual earnings-- it excludes from the actual number any item excluded by the majority of the analysts. However, because the number of contemporaneous analysts cash flow forecasts is much smaller than that of their earnings forecasts (a median over all years of 6 versus 11, as reported in table 1) and the very idiosyncratic nature of the exclusions by different analysts, implementing the above majority rule may result in an I/B/E/S measure of actual cash flows that is not comparable with a significant proportion of the individual forecasts. Further, in many cases, I/B/E/S actual numbers are unavailable. Table 4 provides statistics on the availability of actual cash flow forecasts data by source and the measured

17

A similar lack of uniformity in the definition of the forecasted cash flows is reported by Cao, Wasley and Wu (2007) for management cash flow forecasts.

19

accuracy of analysts cash flow forecasts at the beginning- and end-of-the-year. Since most of the conclusions for the two groups of forecasts are quite similar, we focus on the results in panel B that pertain to the end-of-year forecasts. There are a total of 10,884 firm-years for which consensus end-of-year forecasts could be constructed and actual cash flow numbers were available. For only 5,748 (52.8%) of these firm-years is the actual cash flow number provided by I/B/E/S. For the remaining 5,136 (47.2%) firm-years, only Compustat actual cash flow amounts are available. There are 3,493 firm-years for which actual cash flow numbers are available from both sources. Only in 123 of these cases are the I/B/E/S and Compustat actual amounts identical (see note d to the table). That is, a discrepancy exists between the actual values provided by I/B/E/S and by Compustat (GAAP) in over 95.5% of the cases. This discrepancy arises because Compustat provides the reported cash flow from operations under GAAP whereas the I/B/E/S definition may differ from GAAP due to exclusions of various (generally non-recurring) items. Our finding that the discrepancy is present in 95.5% of the cases is much higher than that documented by past studies of GAAP earnings as compared with I/B/E/S earnings. Dolye et al. (2004) find that the discrepancy due to exclusions of certain items from analysts earnings forecasts (and hence I/B/E/S) is about 20%. Abarbanell and Lehavy (2007), for a slightly different time period, document that the discrepancy in the actual earnings provided by Compustat and I/B/E/S affects about 48% of the firms, still considerably lower than the discrepancy for the cash flow data. The magnitude of the discrepancy for our sample firms (not tabulated) is quite large. The mean (median) absolute difference between the two actual cash flow values deflated by the absolute value of the Compustat reported amount is 0.534 (0.221). Given the discrepancy between the actual Compustat and I/B/E/S values, our use of the Compustat actual amount to define the accuracy of the cash flow forecast when the I/B/E/S actual amount is missing has the potential of biasing upwards the measured forecast error. Indeed, as table 4 shows, for firm-years where I/B/E/S actual amounts are available, their use rather than use of the Compustat actual numbers results in lower forecast errors. For example, for

20

the beginning-of-the-year forecasts reported in panel A (lines 5 and 6), the mean (median) squared error when I/B/E/S is used is 0.797 (0.109) as compared with 1.548 (0.179) when actual values are taken from Compustat. However, even for the cases where I/B/E/S actuals are used to determine the cash flow forecast error, that error is still higher than the error of analysts earnings forecasts. As the last line of panel A shows, the mean (median) squared error of analysts earnings forecast is 0.426 (0.037). While this pattern is also apparent for the median absolute error, it is not present for the mean error. The finding that earnings forecast errors are still considerably smaller than the corresponding cash flow forecast errors even for companies for which I/B/E/S actual data are present is quite pronounced for both error measures for end-ofyear forecasts (see panel B). This suggests that our use of the Compustat actual number for cases where the I/B/E/S actual number is unavailable does not fully explain the higher forecast error of cash flow forecasts relative to earnings forecasts. Focusing on firm-years for which I/B/E.S actual amounts are available allows us to control for both data availability and variability in the underlying series when comparing the accuracy of cash flow forecasts to the accuracy of earnings forecasts. After imposing these controls, we find that the difference in accuracy between cash flow and earnings forecasts made at both ends of the year cannot be explained by the greater variability of the cash flow series. Specifically, the ratio of the median square error of the beginning-of-the-year cash flow forecasts to their contemporaneous earnings forecasts is about 3 (0.109/0.037; panel A lines 5 and 7), lower than the median ratio of the variances of the cash flow. This is particularly true for analysts cash flow forecasts at year end. Specifically, the ratio of the median squared error of cash flow forecasts made at the end of the year to their contemporaneous earnings forecasts for cases where the I/B/E/S actual number is used is about 44 (0.044/0.001; panel B, lines 5 and 7). This value is much higher than the median or mean ratio of the variances of the underlying series. This finding also suggests that there is some deterioration in the relative accuracy of cash flow forecasts as compared with earnings forecasts as the year progresses. The degree of improvement in the cash flow forecasts is more directly examined in section 5.2.3 below.

21

5.2.2. Bias and efficiency The extent of bias in analysts cash flow forecasts, their earnings forecasts, and the timeseries model is shown in Table 5.18 The table confirms the optimistic bias in analysts earnings forecasts at the beginning of the year. The mean bias, measured as the difference between the actual and forecasted earnings amount deflated by the absolute value of actual earnings (or by price) is -0.3055 (-0.2649) (see line 2). This bias essentially disappears by the end of the year. Interestingly, the median bias in analysts earnings forecasts is close to zero at the beginning of the year with the errors distributed almost evenly around zero. The median error becomes slightly more positive (reflecting more pessimistic forecasts or attempts by firms to manage earnings to beat expectations) at year-end, with 58% of the errors being positive (see line 5). Analysts cash flow forecasts suffer as well from a beginning-of-year bias (mean error of -0.2486). Yet, in contrast to the earnings forecasts, a large proportion of this bias still remains at year end as evidenced by the mean error of -0.1356. This may indicate less frequent updating of cash flow forecasts relative to earnings forecasts, consistent with the descriptive statistics on the number of earnings versus cash flow forecasts produced in table 1 and the findings on the intrayear forecast improvement reported in the next section. The time-series model exhibits a much greater bias than that present in analysts forecasts, with a beginning- (end-) of-year mean error of -0.4467 (-0.4444). The results of the efficiency tests (not tabulated) indicate that both cash flow and earnings forecasts exhibit a significant serial correlation. However, the serial correlation between successive cash flow forecast errors is larger than that between successive earnings forecast errors (0.145 vs. 0.096). This finding suggests that cash flow forecasts are less efficient than earnings forecasts and could be improved by a proper adjustment for past errors.

18

Observations in this analysis had all three forecasts available for a given firm-year. Similar results are obtained when all observations available for each model are examined separately.

22

5.2.3. Intra-year improvement Table 6 shows the rate of reduction in the forecast error over the forecast year, measured from the beginning-of-year to the end-of-year forecast. The rate of improvement reflects, among other things, the resources invested by analysts over the year in predicting the outcome as well as the quality of the data. As the table shows, the frequency and rate of improvement are much higher for the earnings forecasts than for the cash flow forecasts. The accuracy of cash flow forecasts improved over the year for 63.6% (64.0%) of the cases when considering the absolute (squared) errors whereas the corresponding percentage of firms with an improvement in their earnings forecasts is 83.5% (83.6%). Further, the median rate of improvement in cash flow forecasts over the forecast year is 0.250 (absolute error) and 0.448 (squared error), much than the lower median improvement rates than for the earnings forecasts, which are 0.756 and 0.941, respectively. These results are consistent with the overall lower quality of cash flow forecasts and suggest that analysts take less care with, and invest fewer resources in, improving their cash flow forecasts during the forecast period as compared with their earnings forecasts. The lack of improvement over the year in the cash flow forecasts is also consistent with the finding reported in section 5.2.1 that, after controlling for other factors dampening the accuracy of cash flow forecasts (namely, the more limited availability of I/B/E/S actual amounts and the higher variability of the cash flow series relative to the earnings series), cash flow forecasts are of a lower accuracy than earnings forecasts. 5.3. Sophistication of analysts cash flow forecasts Table 7 compares the end-of-year forecast errors of the nave earnings-forecast-based model expressed in equation (7) with those of analysts forecasts of cash flows. As the table shows, depending on the error measure considered, the mean difference in the errors is not always in favor of the analysts and when it is, it is either small (e.g., the mean absolute error deflated by the absolute value of the actual cash flow amount is 0.506 for analysts and 0.544 for

23

the naive model) or insignificant, although the median difference under all four error measures exhibit significantly greater accuracy of analysts relative to the nave forecasts of cash flow. Table 8 contains the correlation coefficients between the forecast errors of analysts cash flow forecasts, analysts earnings forecasts and the nave cash flow forecast model (based on analysts earnings forecasts). There is a strong positive correlation between the errors of analysts cash flow forecasts and the errors produced by the nave cash flow forecasts (Pearson coefficients of 0.771 and 0.668 for the price- and actual-deflated errors, respectively). Since the nave cash flow forecasts are based on analysts earnings forecasts, the high correlation between these forecast errors and analysts cash flow forecasts could be attributed to some commonality in the errors in forecasting earnings and cash flows. However, as shown in table 8, the correlation between the errors in analysts cash flow and earnings forecasts is considerably lower (Pearson coefficients of 0.047 and 0.024 for the price- and actual-deflated errors, respectively), suggesting that the high correlation between analysts earnings and cash flow forecasts arises because analysts cash flow forecasts are a nave extrapolation of their earnings forecasts. Table 9 shows the results from estimating regression (8) in which the analysts cash flow forecast is the dependent variables and the independent variables are analysts earnings forecasts, depreciation and amortization, net change in working capital accounts and other adjustments to income made in deriving cash flow from operations. We estimate the regression using alternately the beginning-of-the-year and the end-of-the-year cash flow (and earnings) forecasts. The values of the adjustments to income (e.g., depreciation and amortization) are the realized values for the forecast year.19 If analysts cash flow forecasts are consistent with their earnings forecasts and if, further, analysts consider correctly the various adjustments to net income needed to compute cash flow from operations, we would expect all independent variables to have a positive and significant coefficient that is not significantly different from 1.0. Panel A of the table shows the results

19

We implicitly assume perfect foreknowledge of these realized values since they are unknown to the analyst at the time the forecast is made. Estimation of regression (8) using the recent years values for these variables produces very similar results, suggesting that the above assumption is not critical for our findings.

24

using the beginning-of-the-year forecasts. The cash flow forecasts are very consistent with the earnings forecasts as indicated by the coefficient on the earnings forecasts which is very close to 1.0 and significant. This is also true for the coefficient for depreciation and amortization, suggesting that in producing their cash flow forecasts, analysts add back depreciation and amortization to their earnings forecasts. The coefficients for the net change in working capital and the other adjustments are also significant and positive. However, their value (0.06) is far below 1.0, the level that would be expected if analysts fully and correctly incorporated these adjustments. These low coefficients are consistent with analysts either largely ignoring these adjustments in generating their cash flow forecasts or significantly under-adjusting for these factors. The adjusted R2 values convey the same message. Analysts beginning-of-year earnings forecasts explain 50.7% of the variability in their cash flow forecasts. Adding depreciation and amortization to the regression increases the adjusted R2 significantly, to 76.5%. In contrast, the addition of the working capital and the other adjustments leaves the explanatory power of the regression virtually intact. The same results are obtained when the end-of-year forecasts are used in the regression. These results, tabulated in panel B of table 9, show that even at year-end, analysts either poorly predict the forthcoming change in working capital or the magnitude of other adjustments, grossly understating them, or otherwise do not incorporate this information in their forecasts efficiently. The results from this analysis reinforce the notions that analysts cash flow forecasts represent a simple extension of their earnings forecasts and that the quality of these forecasts (as captured in this analysis by the incorporation of the adjustments to net income) does not improve over the forecast period, consistent with the results reported earlier in section 5.2.3. The finding that analysts cash flow forecasts provide little incremental information beyond their earnings forecasts suggests that investors can easily replicate these forecasts. The finding by previous research that analysts produce these forecasts in response to investors demand (e.g., DeFond and Hung (2003)) may still be valid. Our results only suggest that while

25

analysts respond to what they perceive to be investor demand, the type of cash flow forecasts they produce does not fully meet this demand. Another finding by previous research, namely that the existence of a cash flow forecast affects management reporting behavior (e.g., Collins and McInnis (2006)) remains plausible even in the face of the low incremental information contained in these forecasts. By their mere presence, cash flow forecasts, regardless of their inherent quality, draw investors attention to them and may influence management to consider them in determining their reporting policy because, being in the public domain, these forecasts serve as an additional benchmark against which the reported results might be evaluated. 5.4. Analysts forecasts as a surrogate for market expectation of cash flows Using the association between the forecast error and stock price movements as a measure of the information content of the forecast, we compare the information content conveyed by the analysts consensus cash flow forecasts made at the beginning of the year and that of a mechanical model. Panel A of table 10 presents the results when cumulative abnormal returns over the year are regressed on the beginning-of-year cash flow and earnings forecast errors deflated by price. The results show that the coefficient on FE(CF) is positive and significant, indicating that, after controlling for unexpected earnings, analysts cash flow forecasts have information content. However, the incremental information content of cash flow forecasts beyond that of earnings forecasts is marginal. The adjusted R2 of the regression increases from 5.31% to 6.06% when the cash flow forecast error is included as an additional explanatory variable to the earnings forecast error. While analysts cash flow forecasts appear to provide a greater explanatory power with respect to stock returns relative to the predictions of the mechanical time-series model (an adjusted R2 of 3.20% vs. 1.98%), the incremental information of these forecasts beyond that provided by the earnings forecasts and the mechanical models forecasts is marginal (increasing the adjusted R2 from 5.01% to 6.06%). Interestingly, even in the presence of the earnings and cash flow forecast error variable, the time-series forecast error has a

26 significant coefficient although the increase in explanatory power is slight (the adjusted R2 increases form 6.06% to 6.21%). The results regarding the association between the abnormal return in the short window of the earnings announcement return and end-of-year forecast errors are presented in panel B of table 10. They show that end-of-year analysts cash flow forecasts are not significantly correlated with stock returns and, in particular, do not have incremental information content beyond that provided by end-of-year earnings forecasts. . These results suggest that, after controlling for earnings information, cash flows information (in the form of forecast errors from either analysts or the time-series model) is of marginal value. They also suggest that while analysts forecasts appear to be more aligned with investors expectations, they are almost on par with the time-series model we test. 5.5. Analysts forecasts of cash flows as an accruals expectation model As explained in section 3.4, we examine the effectiveness of an analysts-based accrual expectation model (whereby expected accruals are inferred from the difference between analysts earnings and cash flow forecasts) to detect earnings management with that of two competing models - the modified Jones model and the Dichev and Dechow model. The difference between the expected accruals produced by each model and realized accruals is denoted as unexpected accruals. We use these values of unexpected accruals to determine the presence of earnings management in two subsamples of observations one where earnings management is likely to have occurred and the rest of the sample. As noted earlier, three groups comprise the likely earnings management observations firm-years in which there was a downward restatement of earnings, firm-years where there was a small profit (just-above zero earnings) or small earnings increase (just-above-zero earnings changes). The results concerning the ability of this analysts-based model to identify earnings management are presented in table 11. Recall that earnings management is assumed to exist whenever unexpected accruals exceed 1.5 times their cross-sectional standard deviation.20

20

Similar results are obtained when we use 1.0 or 2.0 standard deviations as demarcation points.

27

Earnings management is considered to have taken place in firm-years with small profits (0 EPS 0.02), small earnings increases when compared with the same quarter in the prior year (0 EPS 0.01), or for which earnings are subsequently restated downward. (For the sake of brevity, we do not tabulate the results for the small earnings increases sample since they are quite similar to the small profits sample.) Several findings emerge from the table. First, the performance of all three models is fairly poor. To illustrate, consider the case were earnings management is presumed to exist whenever earnings are just above zero, 0 EPS 0.02. There are 1,121 such firm-years (out of the 37,067 firm-years for which expected accruals could be computed see panel A of table 11). The modified Jones model identifies only 94 of these cases as earnings management (based on their positive abnormal accruals). This model further classifies 1,075 firm-years as earnings management cases when there is no earnings management. Type I and Type II errors for the model are 3% and 91.6%, respectively (the shaded figures in the table). These errors are only slightly lower than those generated from a nave detection model where a prediction of earnings management for a given firm-year is made randomly with a probability of p and the prediction of no earnings management is made with a probability of (1-p), where p is the proportion of earnings management in the population. This model would produce a Type I error of 3.0% and a Type II error of 97.0% (see table). The same low predictive ability of all three models is observed when earnings managements existence is determined by the presence of a small earnings increase (as noted above, these are not tabulated) or by the presence of a restatement (panel B). The second result from the table is that the analysts-based accrual model does not perform better than the mechanical model and, in fact, performs even worse. This suggests that the measure of unexpected accruals derived from the difference between analysts earnings forecasts and their own cash flow forecasts does not effectively detect earnings management. Our conclusions from this detection ability analysis, however, are subject to a number of caveats. First, in our methodology, earnings management is detected through the presence of

28

sufficiently large unexpected accruals. Yet, earnings management could be achieved with smaller accruals (this is particularly true for earnings management to beat a threshold). Second, our method of identifying likely earnings management cases as those with a small profit, a small earnings increase or a downward restatement, while common in the accounting literature, may lead us to identify innocuous cases as earnings management cases. This is particularly true when the identification is based on small profits or small earnings increases. Indeed, the notion that earnings management is likely to exist when earnings are just above an earnings threshold (e.g., Hayn (1995), Burgstahler and Dichev (1997)) has recently been debated in the literature.21 Finally, in assessing the performance of the analyst-based accrual model, one should consider the possibility that analysts forecasts of earnings subsume analysts expectations of earnings management. That is, the unexpected accruals derived as the difference between analysts earnings forecasts and their cash flow forecasts may exclude earnings-managementinduced accruals. For this reason, the results are consistent with analysts incorporating anticipated earnings management in their earnings forecasts. 6. Concluding Remarks This study examines the quality of cash flow forecasts, an emerging new product of the analysts industry that is currently produced for over 50% of the firms. While the greater frequency of cash flow forecasts appears to be driven by investor demand (DeFond and Hung (2003)), we find that these forecasts are of a considerably lower quality than earnings forecasts. Specifically, cash flow forecasts are much less accurate and are less frequently revised than are earnings forecasts. Further, they appear to involve little more than a nave extension of the accompanying earnings forecasts, leading us to conclude that the difference between the forecasted earnings and cash flows is not a good estimate of the accrual amount expected by investors.

21

See Beaver, McNichols and Nelson (2004), Dechow, Richardson and Tuna (2003), Durtschi and Easton (2005) and Jacob and Jorgensen (2007).

29

There are indications that the low quality of these forecasts is due in part to the presence serious data quality issues. In many instances, the forecasted cash flow variable is defined differently than the actual cash flow variable. Related to this difference is our finding of a high frequency of cases with a discrepancy between the actual cash flow amount reported by I/B/ES and that reported by Compustat. These discrepancies likely contribute to the documented higher prediction errors of analysts forecasts as well as to their poor performance as a proxy for expected accruals. Even though this is an issue of data quality rather than inherent forecast quality, the two are inseparable. Neither investors nor researchers are capable of adjusting the reported cash flows so as to make them consistent with, and therefore a meaningful reference point to, any given cash flow forecast. One possible explanation for the lower accuracy of cash flow forecasts relative to earnings forecasts is that earnings are more likely than cash flows to be managed by the reporting firms to meet analysts forecasts, resulting in lower forecast errors for earnings. This explanation applies primarily to forecasts made late in the year. Our results, however, show that the low relative accuracy of cash flow forecasts prevails throughout the forecasting year. Two comments on the accuracy results are perhaps in order. First, while accuracy is not the only dimension of usefulness (e.g., the forecasts can also assist in interpreting other financial information), it is difficult to envision situations where the presence and content of grossly inaccurate forecasts would help investors. Second, the fact that cash flow forecasts are not universal may raise the issue of self selection whereby analysts cash flow forecasts are demanded ( (and hence supplied) in situations where the prediction of cash flows is difficult and not trivial which may explain the observed high forecast errors for the available cash flow forecasts. Note however that cash flow forecasts are now available for the majority of firms, making the self selection less compelling. Further, our (untabulated) finds show that the accuracy of analysts cash flow forecasts relative to earnings has not improved over time as cash flow forecasts has become more widespread.

30

The finding of the low accuracy of cash flow forecasts, the minor improvement in that accuracy during the year, the fact that they represent in essence a trivial extension of analysts earnings forecasts and the evidence on the presence of serious data quality issues are all relevant for investors who might consider these forecasts in valuation and investment decisions. They are also relevant in research settings where analysts cash flow forecasts are used to proxy for investors expectations or their mere presence serves as an indicator variable (e.g., for earnings quality). The findings also show that analysts cash flow forecasts have an incremental power in explaining contemporaneous annual stock returns (beyond earnings forecasts or a time-series cash flow model). However, this incremental power is marginal, suggesting that in certain research settings researchers may use time-series models instead of analysts cash flow forecasts without adversely affecting the power of the tests. Finally, while not invalidating their role as an indicator of earnings quality and of investor demand, the finding that these forecasts are of low quality and in essence a nave extension of the earnings forecasts suggests that a better understanding is needed of the source of the signaling value associated with the presence of cash forecasts.

31 References Abarbanell, J. and R. Lehavy, 2003a, Can Stock Recommendations Predict Earnings Management and Analysts Earnings Forecast Errors? Journal of Accounting Research, v. 41 (1), 1-31. __________ and __________, 2003b, Biased Forecasts or Biased Earnings? The Role of Reported Earnings in Explaining Apparent Bias and Over/Underreaction in Analysts Earnings Forecasts, Journal of Accounting and Economics, v. 36 (1-3), 105-146. __________ and __________, 2007, Letting the Tail Wag the Dog: The Debate over GAAP versus Street Earnings Revisited, Contemporary Accounting Research, Fall, 657-74. Ball, R. and L. Shivakumar, 2005, Earnings Quality in U.K. Private Firms: Comparative Loss Recognition Timeliness, Journal of Accounting and Economics, v. 39(1), 83-128. Barth, M.E., Cram, D. and K. Nelson, 2001, Accruals and the Prediction of Future Cash Flows, The Accounting Review, v. 76, 27-58. Brown, L.D., Hagerman, R.M. and M.E. Zmijewski, 1987, Security Analyst Superiority Relative to Univariate Time-Series Models in Forecasting Quarterly Earnings, Journal of Accounting and Economics, v. 9, 61-87. _________ and A.S. Pinello, 2008, Why Do Firms Meet or Beat Analysts Cash Flow Forecasts but Miss Their Earnings Forecasts? Working Paper, Georgia State University. Burgstahler D. and I. Dichev, 1997, Earnings Management to Avoid Earnings Decreases and Losses, Journal of Accounting and Economics, v. 24, 99-126. ____________ and M. Eames, 2003, Earnings Management to Avoid Losses and Earnings Decreases: Are Analysts Fooled? Contemporary Accounting Research, Summer, 253294. Call, A. C., 2007, The Implications of Cash Flow Forecasts for Investors Pricing and Managers Reporting of Earnings, Working Paper, University of Washington. Cao, Y., Wasley, C.A.E. and J.S. Wu, 2007, Soft-Talk Management Cash Flow Forecasts: Verifiability, Credibility and Stock Price Effects, Working Paper, University or Rochester. Dechow, P. and I. Dichev, 2002, The Quality of Accruals and Earnings: The Role of Accrual Estimation Errors, The Accounting Review, v. 77 (Supplement), 35-59. _________ , Richardson, S.A. and A.I. Tuna, 2003, Why are Earnings Kinky? An Examination of the Earnings Management Explanation, Review of Accounting Studies, v. 8, 355-384. _______, Kothari, S.P. and R. Watts, 1998, The Relation between Earnings and Cash Flows, Journal of Accounting and Economics, v. 25, 133-168.

32 __________, Sloan, R. and A. Sweeney, 1995, Detecting Earnings Management, The Accounting Review, v. 70, 193-225. DeFond, M. and M. Hung, 2003, An Empirical Analysis of Analysts Cash Flow Forecasts, Journal of Accounting and Economics, v. 35, 73100. __________ and ________, 2007, Investor Protection and Analysts Cash Flow Forecasts Around the World, Review of Accounting Studies, v. 12, 377-419. Desai H., Krishnamurthy, S. and K. Venkataraman, 2006, Do Short Sellers Target Firms with Poor Earnings Quality? Evidence from Earnings Restatements, Review of Accounting Studies, v. 11, 71-90. Dichev, I. D. and V. W. Tang, 2007, Earnings Volatility and Earnings Predictability, Working Paper, University of Michigan. Doyle, J.T., McNichols, M.F. and M.T. Soliman, 2004, Do Managers Define Street Earnings to Meet or Beat Analyst Forecasts? Working Paper, Stanford University. Durtschi, C. and P. Easton, 2005, "Earnings Management? The Shapes of the Frequency Distributions of Earnings Metrics are not Evidence Ipso Facto," Journal of Accounting Research, v. 43, 557-592. Ertimur, Y. and S. Stubben, 2005, Analysts Incentives to Issue Revenue and Cash Flow Forecasts, Working Paper, Stanford University. Ettredge, M., Shane, P. and D. Smith, 1995, "Overstated Quarterly Earnings and Financial Analysts' Earnings Forecast Revisions," Decision Sciences, November/ December, 781-801. Fried, D. and D. Givoly, 1982, "Financial Analysts' Forecasts of Earnings: A Better Surrogate for Market Expectations," Journal of Accounting and Economics, v. 4, 85-107. Givoly, D. and J. Lakonishok, 1979, The Information Content of Analysts Forecasts of Earnings, Journal of Accounting and Economics, v. 1, 165-185. _______ , Hayn, C. and T. Yoder, 2007, What Do Analysts Really Predict: Evidence from Earnings Restatements and Managed Earnings, Working Paper, Pennsylvania State University. Hail, L., 2007, Discussion of Investor Protection and Analysts Cash Flow Forecasts Around the World, Review of Accounting Studies, v. 12, 421-441. Hayn, C., 1995, The Information Content of Losses, Journal of Accounting and Economics, v. 20, 125-153. Jacob, J. and B. Jorgensen, 2007, Earnings Management and Accounting Income Aggregation, Journal of Accounting and Economics, v. 43(2-3), 369-390.

33 Jones, J., 1991, Earnings Management During Import Relief Investigations, Journal of Accounting Research, v. 29, 193-228. Liu, X., 2005, Analysts Response to Earnings Management, Working Paper, University of Texas at Dallas. McInnis, J. and D. W. Collins, 2006, Do Cash Flow Forecasts Deter Earnings Management? Working Paper, Tippie College of Business, University of Iowa. Melendrez, K.D., Schwartz, W.C. and M.A. Trombley, 2005, How Does the Market Value Accrual and Cash Flow Surprises? Working Paper, University of Arizona. Pae, J., Wang, S. and C. Yoo, 2007, Do Analysts Who Issue Cash Flow Forecasts Predict More Accurate Earnings? Working Paper, Queens University. Zhang, W., 2008, The Effects of Meeting or Beating Analysts Cash Flow Forecasts on Market Reaction and Analysts Forecast Revisions, Working Paper, University of Texas at Dallas. Wasley C.A.E. and J.S. Wu, 2005, Why Managers Voluntarily Issue Cash Flow Forecasts? Working Paper, University or Rochester.

This table presents descriptive statistics on the availability of cash flow and earnings forecasts in the I/B/E/S Detail data files. Panel A indicates the total number firms for which a minimum of one one-year-ahead forecast of cash flows or earnings, respectively, was made during the year. Data on the mean and median number of forecasts per firm as well as the number of issuing analysts is also provided. Panel B shows how many forecasts were received by firms over time. Panel C provides the I/B/E/S sector classification for the firms with valid cash flow and earnings forecasts.

Panel A. By Year

Year Number of Firms with at Least One One-YearAhead Forecast Made During the Year (1) CF Forecasts (2) Earnings Forecast (3) = (1) / (2) % of Firms with a CF Forecast Number per Firm of: Analysts Issuing Earnings Cash Flow Forecasts Forecasts Mean Median Mean Median Analysts Issuing Earnings Forecasts Mean Median

Year 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 All Years

118 579 811 799 965 1,131 1,715 1,689 1,173 1,830 2,504 2,947 3,261 19,522

4,672 5,193 5,684 6,378 6,796 6,853 6,587 6,132 5,129 4,890 4,917 5,456 5,706 74,393

2.5% 11.1% 14.3% 12.5% 14.2% 16.5% 26.0% 27.5% 22.9% 37.4% 50.9% 54.0% 57.2% 26.2%

3.2 7.6 8.0 15.7 15.1 15.5 14.1 12.8 13.9 10.2 13.9 14.1 14.3 13.3

2 2 4 8 8 7 5 6 3 3 7 7 7 6

3.1 3.5 3.7 5.6 5.2 5.0 4.0 3.9 4.4 3.3 3.7 3.5 3.3 3.9

2 2 2 4 3 3 2 2 2 2 2 2 2 2

21.6 19.8 20.0 18.8 18.4 19.9 20.3 20.5 25.3 25.1 26.8 27.3 28.0 22.2

12 11 11 10 10 10 10 10 12 13 14 14 15 11.0

7.3 6.9 6.6 6.3 6.2 6.3 6.6 6.8 7.6 7.5 7.6 7.2 7.3 6.9

4 4 4 4 4 4 4 4 5 5 5 5 5 4.0

35 Table 1 (continued) Descriptive Statistics on Availability of Cash Flow Forecasts Panel B: By Industry Group

Number of Firms with at Least One One-Year-Ahead Forecast Made During the Year

Industry Group

(Based on I/B/E/S Industry Group Classification)

CF Forecast

1995

% of Firms with a CF Forecast a

2005

CF Earnings Forecast Forecast % of Firms with a CF Forecast a

Basic Industries Capital Goods Consumer Durables Consumer Non-Durables Consumer Services Energy Finance Health Care Public Utilities Technology Transportation Miscellaneous/ Undesignated All Industries

a

3,220 1,306 607 1,020 3,034 3,998 1,590 1,187 1,221 1,654 502 190 19,522

6,940 6,256 2,792 3,799 11,857 5,083 12,359 8,076 3,376 12,001 1,415 439 74,393

46.4% 20.9% 21.7% 26.8% 25.6% 78.7% 12.9% 14.7% 36.2% 13.8% 35.5% 43.3% 26.2%

626 540 273 338 883 338 904 560 51 275 781 114 5,684

30.4% 10.4% 11.0% 15.1% 11.8% 66.3% 2.8% 5.0% 51.0% 17.5% 2.2% 10.5% 14.3%

439 213 111 153 505 461 367 289 23 187 412 101 3,261

555 396 180 253 839 476 1064 680 31 230 873 129 5,706

79.1% 53.8% 61.7% 60.5% 60.2% 96.8% 34.5% 42.5% 74.2% 81.3% 47.2% 78.3% 57.1%

% of firms with an earnings forecast that also have a cash flow forecast.

36 Table 2 Comparison of the Accuracy of Analysts Cash Flow Forecasts, Analysts Earnings Forecasts and Cash Flow Forecasts based on the Time-Series Model

This table reports statistics on the accuracy of analysts cash flow forecasts relative to that of analysts earnings forecasts and forecasts generated from a timeseries model. Accuracy is defined as the absolute and the squared values of the forecast error. The absolute (squared) forecast error is computed as the absolute (squared) value of the difference between the actual and the forecasted number, scaled by the absolute value of the actual number. The cash flow forecast generated by the time-series model is based on the regression of: CFt = 0 + 1CFt-1+2 ARt-1 + 3INVt-1 + 4APt-1+5DEPt-1+6Othert-1 + t. Statistics in Panel A (Panel B) report comparisons based on the consensus forecast issued at the beginning (end) of the year. The analysis in this table is restricted to observations for which all three measures are available. All variables are truncated at the top percentile.

Type of Forecast A. Beginning-of-Year Forecasts (n=4,766)a (1) Analysts CF Forecasts (2) Analysts Earnings Forecasts (3) Time-series Model

Difference: (1) (2) (p-value) Difference:(1) (3) (p-value)

0.0433 (0.0827) -0.1350 (<0.0001)

0.0796 (<0.0001) 0.0045 (0.4609)

0.5306 (<0.0001) 0.1002 (0.0533)

0.1149 (<0.0001) 0.0091 (0.2348)

B. End-of-Year Forecasts (n=6,817)a (1) Analysts CF Forecasts (2) Analysts Earnings Forecasts (3) Time-series Model

Difference: (1) (2) (p-value) Difference:(1) (3) (p-value)

0.3595 (<0.0001) -0.2426 (<0.0001)

0.1725 (<0.0001) -0.0593 (<0.0001)

0.7189 (<0.0001) -0.0581 (<0.0001)

0.0784 (<0.0001) -0.0670 (<0.0001)

Observations included in must have all three forecasts available. Relative Absolute Error =Ait Fit / Ait c Relative Squared Error = (Ait Fit)2 /Ait

b

37

Table 2 (cont.) Comparison of the Accuracy of Analysts Cash Flow Forecasts, Analysts Earnings Forecasts and Cash Flow Forecasts based on the Time-Series Model C. Beginning-of-Year Forecasts, by Year

Analysts Cash Flow Forecasts Year 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 By Subperiod: 1993-1999 2000-2006 2,999 7,553 0.405 0.524 0.119 5.43 0.119 0.219 0.100 15.10 0.627 0.934 0.307 0.022 0.086 0.267 0.142 -0.125 -10.91 0.078 0.036 0.093 0.044 0.004 0.001 N 72 303 459 499 530 472 664 788 434 1,044 1,609 1,775 1,903 Relative Absolute Error Mean Median 0.369 0.104 0.394 0.103 0.228 0.082 0.291 0.085 0.359 0.103 0.409 0.140 0.659 0.250 0.596 0.233 0.644 0.138 0.528 0.221 0.526 0.222 0.480 0.237 0.504 0.212 Relative Squared Error Mean Median 0.490 0.013 0.523 0.017 0.313 0.011 0.430 0.012 0.393 0.017 0.673 0.031 1.208 0.103 1.197 0.115 1.103 0.037 0.941 0.075 0.871 0.091 0.768 0.090 0.992 0.086 Analysts Earnings Forecasts Relative Absolute Error Mean Median 0.260 0.116 0.276 0.100 0.288 0.086 0.231 0.076 0.280 0.071 0.288 0.090 0.249 0.059 0.164 0.042 0.189 0.040 0.143 0.035 0.132 0.034 0.140 0.035 0.132 0.038 Relative Squared Error Mean Median 0.110 0.006 0.088 0.005 0.103 0.005 0.076 0.004 0.078 0.003 0.114 0.005 0.099 0.003 0.049 0.002 0.076 0.001 0.036 0.001 0.041 0.001 0.045 0.001 0.042 0.002

0.064 16.80

0.064 -15.90

-0.049 -6.31

-0.003 -13.20

4.95

38 Table 3 Distribution of the Variance of the Time-Series of Cash Flows and Earnings

This table reports statistics on the variance of the cash flow per share and earnings per share series, standardized either by assets per share or price at the end of the year. Firms must have at least one cash flow forecast and 10 years of data on actual cash flow and earnings to participate in the analysis. All variances are multiplied by 1,000.

Variance

Standardized by Assets (n = 1,533) Cash flow per share Earnings per share Ratio of Cash Flow Variance to Earnings Variance 7.02 13.75 5.69 1.01 0.58 0.56 2.53 1.85 1.18 6.35 7.35 2.56 56.8%

Standardized by Price (n = 1,359) Cash flow per share Earnings per share Ratio of Cash Flow Variance to Earnings Variance 72.56 380.8 8.62 1.04 0.38 0.63 3.11 1.39 1.81 10.56 10.26 4.44 65.6%

39 Table 4 Accuracy of Analysts Cash Flow Forecasts Conditional on Availability of Actual Cash Flow Numbers

This table reports statistics on the accuracy of analysts cash flow forecasts for different subsamples. Accuracy is defined as the absolute and the squared values of the forecast errors. Absolute (squared) forecast errors are computed as the absolute (squared) value of the difference between the actual and the forecasted number, scaled by the absolute value of the actual number. All variables are truncated at the top percentile.

Observations for which the actual cash flow amount is available: on I/B/E/S only on Compustat but not on I/B/E/S only on I/B/E/S on both I/B/E/S and Compustatd - Errors are based on I/B/E/S actual amount - Errors are based on Compustat actual amount - Earnings forecast errors are based on I/B/E/S actual amount No. of Obs. Absolute Forecast Errora

Mean Median

Mean Median

7,543c

0.603

0.258

1.143

0.138

2 3 4 5 6 7

No. of Obs.

Mean Median

Mean Median

All Observations

Observations for which the actual cash flow amount is available:

10,884c 0.506

0.191

0.965

0.064

2 3 4 5 6 7

a c

on I/B/E/S only on Compustat only on I/B/E/S on both I/B/E/S and Compustat - Errors are based on I/B/E/S actual amount - Errors are based on Compustat actual amount - Earnings forecast errors are based on I/B/E/S actual amount

d

3,455

0.140

0.031

0.051

0.001

Absolute Forecast Error =Ait Fit / Ait; b Squared Forecast Error = (Ait Fit)2 / Ait This is the number of firm-years for which (1) consensus beginning-of-the-year cash flow forecasts can be constructed (see section 3) and (2) data on the actual cash flow is available. (Because of this restriction, this number is lower than the number of firm-years with at least one cash flow forecast, 19,522, as reported in table 1.) d Compustat and I/B/E/S actual amounts are identical for only 123 firm-years (or less than 4% of these cases).

40 Table 5 Bias of Analysts Cash Flow Forecasts, Analysts Earnings Forecasts and Cash Flow Forecasts based on the Time-series Modela

This table reports statistics on the bias in analysts cash flow forecasts, analysts cash flow forecasts, and forecasts generated from the time-series model. Bias is defined the realized minus the forecasted value scaled alternatively by the absolute value of the realized value and the price at the beginning of the period. All variables are truncated at the top and bottom percentile.

Time of Forecast

Sign of Forecast Error Line Type of Forecast % Pos. 1 Analysts Cash Flow Analysts Earnings Time-Series Model Analysts Cash Flow Analysts Earnings Time-Series Model 55.6 50.3 51.1 57.8 58.0 51.7 2 3 4 % Neg. 44.2 48.7 48.9 41.7 33.7 48.3 % Zero 0.2 1.0 0.0 0.5 8.3 0.0

Bias deflated by Absolute Actual Value (panel A: n=4,697; panel B: n=6,734) Mean -0.249 -0.306 -0.447 -0.136 -0.020 -0.444 Q1 -0.253 -0.258 -0.348 -0.156 -0.020 -0.362 Median 0.048 0.003 0.018 0.041 0.009 0.018 Q3 0.261 0.135 0.218 0.235 0.045 0.225

Bias deflated by Price (x100) (panel A: n= 4,651; panel B: n=6,680) Mean 0.911 -0.265 -0.866 0.895 -0.029 -0.672 Q1 -2.272 -1.104 -2.568 -1.293 -0.090 -2.480 Median 0.436 0.017 0.081 0.359 0.044 0.125 Q3 3.312 0.836 2.074 2.427 0.226 2.244

Beginningof-Year

End-of-Year

a b

5 6

BIASit = (Ait Fit) / Ait CFit = 0 + 1CFit-1+2 ARit-1 + 3INVit-1 + 4APit-1+5DEPit-1+6Otherit-1 + it (equation 6)

This table reports statistics on the intra-year improvement in analysts cash flow and earnings forecast accuracy. Rate of improvement is defined as the absolute (squared) forecast error based on the consensus forecast at the beginning of the year divided by the absolute (squared) forecast error based on the consensus forecast at the end of the year, minus 1. Absolute (squared) forecast error is computed as the absolute (squared) value of the difference between the actual and the forecasted number, scaled by the absolute value of the actual. All variables are truncated at the top percentile.

Type of Analysts Forecasts Cash Flow Forecasts (n=7,375) Earnings Forecasts (n=9,140)

a

Error Measureb Absolute forecast errors Squared forecast errors Absolute forecast errors Squared forecast errors

Rate of Improvementc Mean -0.500 -9.836 0.394 -0.529 Quartile 1 -0.333 -0.742 0.358 0.587 Median 0.250 0.448 0.756 0.941 Quartile 3 0.713 0.920 0.917 0.993

Table is based on observations for which both beginning-of year and end-of-year earnings and cash flow forecasts are available. b Absolute Error =Ait Fit / Ait Squared Error = (Ait Fit)2 / Ait c Rate of Improvement = 1 - (Errori,END(t) /Errori,BEG(t) )

This table compares the absolute and squared analysts cash flow forecast error to those based on a nave forecast model. Analysts forecast errors are based on the consensus cash flow forecast at the end of the year. The naive forecast errors are based on the following model: F(CFit) = 0 + 1 F(Earningsit) + 2DEP it + it, where F(Earnings) is analyst consensus earnings forecast and DEP is the amount of depreciation (Compustat data item 125 scaled by number of shares). Absolute (squared) forecast errors is computed as the absolute (squared) value of the difference between the actual and the forecasted number, scaled by the absolute value of the actual. All variables are truncated at the top percentile.

Type of Forecast Analysts Cash Flow Forecast Nave Forecast of Cash Flowb

Absolute Errora Mean 0.506 0.544 -0.038* Median 0.191 0.212 -0.021**

Squared Errora Mean 0.965 0.892 0.073 Median 0.064 0.091 -0.031**

* (**) significant at the 5% (1%) significance level under a one-sided hypothesis. a Absolute Error =Ait Fit / Ait Squared Error = (Ait Fit)2 / Ait b The nave forecast model is F(CFit) = 0 + 1 F(Earningsit) + 2DEP it + it

This table reports the Pearson (above the diagonal) and Spearman correlation (below the diagonal) coefficients between the signed errors deflated by the absolute value of the actual amount for analysts forecasts of cash flows, a naive cash flow forecast model, and analysts forecast of earnings. The naive forecast errors are based on the following model: F(CFit) = 0 + 1 F(Earningsit) + 2DEPit + it, where F(Earnings) is analysts consensus earnings forecast and DEP is the amount of depreciation (Compustat data item 125 scaled by number of shares). All variables are truncated at the top percentile.

Analysts Cash Flow Forecast Errors Analysts Cash Flow Forecast Errors Nave Model Forecast Errorsc Analysts Earnings Forecast Errors

a b

The number of observations varies from 6,647 to 10,295 depending on the cell. Signed errors deflated by the absolute value of the actual amount =(Ait Fit) / Ait c The nave forecast model is F(CFit) = 0 + 1 F(Earningsit) + 2DEPit + it

This table reports coefficient estimates and t-statistics (in parentheses) of estimating regression (8): F (CF)t = 0 + 1 F(Earnings) t + 2DEPt + 3WCt + 4OTHERt + t. Analyst earnings forecasts are the consensus forecasts at the beginning (panel A) and end of the year (panel B). Depreciation and amortization (DEP) is Compustat data item 125. Change in working capital (WC) is equal to the sum of Compustat data items 302, 303, 304. Other adjustments, OTHER, is equal to Compustat data item 308 minus the sum of Compustat data items 123, 125, 302, 303, 304. All variables are deflated by average total assets and are truncated at the top and bottom percentiles. (t-statistics are provided in parentheses.)

A. Beginning-of-Year Cash Flow Forecasts (n = 3,719) Coefficients Analysts Earnings Intercept DEP Forecasts 0.907 0.059

(51.8)

WC

Other

(61.9) (93.7)

-0.03

(-2.0)

0.951 0.954

(93.6)

1.045

(63.8)

-0.002

(-1.7)

1.042

(63.7)

0.058

(3.3)

76.6% 0.060

(4.7)

-0.003

(-2.2)

0.952

(93.7)

1.038

(63.5)

0.064

(3.7)

76.7%

B. End-of-Year Cash Flow Forecasts (n = 5,348) Coefficients Analysts Earnings Intercept DEP Forecasts 0.059 0.886

(51.0) (84.8)

WC

Other

0.003

(2.0)

0.911

(109.7)

0.977

(56.1)

0.003

(2.4)

0.916

(109.3)

0.974

(56.0)

0.070

(4.1)

73.2% 0.129

(20.1)

0.001

(0.8)

0.918

(110.6)

0.968

(56.1)

0.086

(5.0)

73.7%

DEP: Depreciation and amortization WC: Change in working capital accounts Other: Other adjustments to income from continuing operations needed to reconcile it with the cash flows from operating activities

This table reports coefficient estimates and t-statistics (in parentheses) of the following regression: CARjt = +1 FE(Earningsit) + 2FE(CFit) + it.

CAR is the cumulative abnormal return for the year beginning 8 months prior to and 3 months after the fiscal yearend. FE(Earnings) is the forecast error computed from the beginning-of-the year earnings forecast deflated by price. FE(CF) is the forecast error computed from the beginning-of-the-year cash flow forecast, deflated by price. The independent variables (the forecast errors) are multiplied by 100 before estimating the regression. N = 4,485. (T-statistics are shown in parentheses.)

Intercept 0.123 (15.6) 0.105 (13.1) 0.12 (15.3) 0.116 (14.3) 0.126 (15.9) 0.119 (14.8)

0.010 (9.6) 0.027 (11.7) 0.031 (13.3) 0.026 (10.8) 0.006 (6.1) 0.006 (5.0) 0.047 (4.4) 0.003 (2.8)

CAR is the cumulative abnormal return for the 4-day period beginning one day prior to the earnings announcement day. FE(Earnings) is the forecast error computed from the end-of-year earnings forecasts deflated by price. FE(CF) is the forecast error computed from the end-of-year cash flow forecasts deflated by price. The independent variables (the forecast errors) are multiplied by 100 before estimating the regression. N = 6,045.

Intercept 0.421 (4.6) 0.402 (4.4) .427 (4.7) 0.407 (4.5) 0.431 (4.7) 0.420 (4.6)

0.025 (2.2) 0.262 (3.9) 0.251 (3.7) 0.251 (3.7) 0.014 (1.22) 0.097 (0.8) 0.021 (1.8) 0.019 (1.6)

(Detection occurs when unexpected accruals > 1.5 standard deviations of the unexpected accruals distribution)

Modified Jones Model Classification A-priori Determination EM exists (0 EPS 2 cents) EM does not exist All firm-years %

3.0 97.0 100.0

Total

1,121 (100.0%) 35,946 (100.0%) 37,067

EM

94 (4.9%) 1,075 (3.0%) 1,169

No EM

1,027 (91.6%) 34,871 (91.6%) 35,898

Improvement over benchmark: Type I error (in sample, 3.0%; from nave benchmark model*, 3.0%) 0.0% Type II error (in sample, 97.0%; from nave benchmark model*, 91.6%) 5.4%

A-priori Determination EM exists (0 EPS2 cents) EM does not exist All firm-years

%

3.0 97.0

1,971 (100.0%) 64,328 (100.0%) 66,299 80 (4.1%) 1,034 (1.6%) 1,114 1,891 (95.9%) 63,294 (98.4%) 65,185

100.0 Improvement over benchmark: Type I error (in sample, 3.0%; from nave benchmark model*, 1.6%) 1.4% Type II error (in sample, 97.0%; from nave benchmark model*, 95.9%) 1.1%

A-priori Determination EM exists (0 EPS 2 cents) EM does not exist All firm-years

%

0.7 99.3

58 (100.0%) 7,880 (100.0%) 7,938 4 (6.9%) 292 (3.7%) 296 54 (93.1%) 7,588 (99.3%) 7,642

100.0 Improvement over benchmark: Type I error (in sample, 0.7%; from nave benchmark model*, 3.7%) 3.0% Type II error (in sample, 99.3%; from nave benchmark model*, 93.1%) 6.2%

*The nave benchmark model predicts earnings management (for a given firm-year) with a probability of p and no earnings management with a probability of (1-p), where p is the proportion of earnings management in the sample.

47 Table 11 (continued) Ability of Alternative Accrual Models to Detect Correctly Earnings Management

(Detection occurs when unexpected accruals > 1.5 standard deviations of the unexpected accruals distribution)

B. Earnings Management is Assumed to Exist When Earnings are Subsequently Restated Downward

Modified Jones Model Classification A-priori Determination EM exists (annual earnings were restated) EM does not exist All other firms in the restatement year that belong to the same 4-digit industry %

2.9 97.1 100.0

Total

120 (100.0%) 3,668 (100.0%) 3,778

EM

2 (1.7%) 60 (1.6%) 62

No EM

118 (98.7%) 3,608 (98.4%) 3,726

Improvement over benchmark: Type I error (in sample, 2.9%; from nave benchmark model*, 1.6%) 1.3% Type II error (in sample, 97.1%; from nave benchmark model*, 98.7%) -1.6%

A-priori Determination EM exists (annual earnings were restated) EM does not exist All other firms in the restatement year that belong to the same 4-digit industry

%

3.4 96.6 100.0

175 (100.0%) 4,966 (100.0%) 5,141 0 (0.0%) 56 (1.1%) 56 175 (100.0%) 4.910 (98.9%) 5.085

Improvement over benchmark: Type I error (in sample, 3.4%; from nave benchmark model*, 1.1%) 2.3% Type II error (in sample, 96.6%; from nave benchmark model*, 100.0%) -3.4%

Analyst-Based Classification A-priori Determination EM exists (annual earnings were restated) EM does not exist All other firms in the restatement year that belong to the same 4-digit industry %

3.4 96.6 100.0

Total

80 (100.0%) 2,280 (100.0%) 2,360

EM

1 (1.3%) 40 (1.8%) 41

No EM

79 (98.7%) 2,240 (98.2%) 2,319

Improvement over benchmark: Type I error (in sample, 3.4%; from nave benchmark model*, 1.8%) 1.6% Type II error (in sample, 96.6%; from nave benchmark model*, 98.7%) -1.1% *The nave benchmark model predicts earnings management (for a given firm-year) with a probability of p and no earnings management with a probability of (1-p), where p is the proportion of earnings management in the sample.

- 1. Forecasting IntroductionUploaded byRajan Manchanda
- WINSEM2018-19_MEE1014_TH_MB102_VL2018195002165_Reference Material I_Unit 1.2_IEM (1).pdfUploaded byDevanshu Sahay
- bb sbi FAN-MANAGING EARNING USING CLASIFICATION SHIFTING.pdfUploaded byTriadiPrasetyo
- 17833356Uploaded byjyottsna
- Sales ForecastingUploaded byvinaycool12344150
- Evaluation of the Quality of a Prognosis for an Industrial Product using the Regression AnalysisUploaded byEditor IJTSRD
- SupplyChainPlanningAndControl BA EnUploaded byrajivbi
- chap_15.pptUploaded byBinh Do
- Gat Subject Management Sciences Management Mcqs1 50Uploaded byNazia Attique
- ForecastingUploaded byAnkkit Pandey
- Operations Management chapter 3 - ForecastingUploaded byGio Mintu
- Demand ForecastingUploaded bymanojkumar024
- 4 Cast DemandUploaded bymanuelmiguel_fbi
- ForecastingUploaded bydinesh
- Solanib_Quanti.pdfUploaded byyy
- Swanson & Tedrow Paa 2016 v1Uploaded byCésar Andrés Cristancho Fajardo
- Demand Forecasting TechniquesUploaded byVishakha Pawar
- 360 case 2 fullUploaded byRafayat Rakib
- Demand and Supply Network Planning in SCM.pdfUploaded bySuresh Reddy
- 170Uploaded bySharmin Sultana
- MB0044Uploaded bySmu Doc
- ThinkVine Boosting Media ValueUploaded byDatalicious Pty Ltd
- ForecastingUploaded bysamrulezzz
- HW#9.pdfUploaded byMartin De Haro Garcia
- Forecasting Lectura Ingles(1)Uploaded byLuis Hernando Herrera
- Forecasting for Asian PaintsUploaded bysampritc
- Project Cost ForecastingUploaded byAjitabh Singh
- 3.0 Business ForecastingUploaded bySaiKiran
- MKT401 Learning ObjectiveUploaded byphanny
- Regional Rainfall ForecastUploaded byNarayana Iyengar

- Quick Start GuideUploaded byAlvarezd
- Gretl User GuideUploaded byÉva Lovas
- Pattern Discovery Fuzzy Time SeriesUploaded byPrasanta Debnath
- SSRN-id488144Uploaded byTrinh Le Tan
- Get to Know the IBM SPSS Product Portfolio.pdfUploaded byImmanuel Henne
- Exercises in Advanced Risk and Portfolio Management.pdfUploaded bydoc_oz3298
- Quantitative+ApplicationsUploaded byshukla_ankit
- A Primer on Cash Flow ForecastingUploaded bysreesarma
- Beck y Katz 2011Uploaded byMaría Marta Maroto
- Blueman 5th_Chapter 2 HW SolnUploaded byAlbert Ameka
- PPA_691_Lecture5-6Uploaded bykuldip singh
- Comparison Of Neural Network And Multivariate Discriminant Analysis In Selecting New Cowpea VarietyUploaded byijcsis
- Annexure a Syllabus BEUploaded byAdnan Husain Shiekh
- Advanced Topics in Time Series Eco No Metrics Using R1_ZongwuCAIUploaded byYang
- forecasting engineeringUploaded byHerbert Ebrada
- Data Preparation: Stationarity (Excel)Uploaded bySpider Financial
- AutocorrelationUploaded byJenab Peteburoh
- Modeling Equilibrium Relationships Error Correction Models With Strongly Auto Regressive DataUploaded byrunawayyy
- Effects of Word-Of-Mouth Versus Traditional Marketing Findings From an Internet SocialUploaded byJavier Gonzalez
- FROM EFFICIENT MARKET THEORY TO BEHAVIORAL FINANCEUploaded byapi-3776228
- Introduction to the Canback Global Income Distribution Database (C-GIDD)Uploaded byIVAN
- Ecological Indicators to Capture the Effects of Fishing on BiodiversityUploaded byJohn Zoof
- Rainfall DataUploaded byYagnesh Dave
- PublicationsUploaded bymukesh3021
- Impact of Macroeconomic Variables on Oil and Gas Stock MarketUploaded byMati Ullah Khan
- Telling Compelling Stories With NumbersUploaded byAndré Lopes
- Lecture Note 10 Adf 2Uploaded byz_k_j_v
- 110105053Uploaded byMarkWeber
- Demand ForecastingUploaded byAnkit Khambhatta
- The Vanishing January EffectUploaded bySyed Muhammad Hasan Bilal